Wall Street Millennial
July 7, 2026
Chinese AI Models Disrupting Enterprise Market with Lower Costs
TL;DR
Chinese open-source AI models like Deepseek and GLM are gaining significant enterprise adoption due to dramatically lower token costs, threatening the business model of U.S. companies like OpenAI and Anthropic despite remaining niche in consumer markets.
Key Takeaways
- 1Chinese AI models dominate enterprise adoption via OpenRouter despite fading from consumer markets; token usage exploded from 2.4 trillion to 47 trillion per week in a year
- 2Z.AI's GLM 5.2 matches Claude and ChatGPT performance while costing 75% less ($1.40 vs $5 per million input tokens), making it attractive for cost-sensitive enterprises
- 3Chinese labs achieve competitive results with under $2 billion in funding versus OpenAI and Anthropic's $100+ billion through model distillation and efficient training
- 4Enterprise customers like Airbnb choose open-source Chinese models for customization and cost savings, while consumer markets remain dominated by ChatGPT and Google Gemini
- 5LLM commoditization threatens U.S. frontier labs' token-based business model as enterprises will inevitably seek cheaper alternatives despite rapid development cycles
Notable Quotes
“GLM 5.2 is about one quarter of the cost of Claude Opus 4.7”
— Wall Street Millennial
“Z.AI developed its business strategy based on the idea that LLMs will become commoditized with a race to the bottom for token prices. The way to make money is with ancillary services like model customization.”
— Wall Street Millennial
“Open AI can spend billions of dollars developing new frontier models, which are ever bigger and more expensive to operate. But within a few months, the Chinese labs are likely to catch up and undercut them on price.”
— Wall Street Millennial
“LLMs are quickly becoming commoditized”
— Wall Street Millennial
Chapters
1. Deepseek Shock and Consumer Market Reality
Deepseek's January 2025 launch shocked markets but consumer dominance tells a different story: ChatGPT retains 1 billion monthly active users while Deepseek peaked at 200 million then declined. Google Gemini sits at 600 million users.
2. Z.AI: The Hidden Winner in Enterprise AI
Z.AI went public in January 2026 with IPO price of 116 HKD per share, reaching 1,800 HKD (100 billion USD market cap) by July. Generated 107 million USD revenue in 2025 through cloud subscriptions (26%) and on-premise deployment (74%), but operated at loss.
3. GLM Models and Competitive Performance
GLM 5.2 released June 2026 matches Claude and ChatGPT on benchmarks, particularly excelling at long-horizon coding tasks. Priced at $1.40/$4.40 per million tokens versus Opus 4.7's $5/$25, offering 75% cost savings with option to run locally.
4. Enterprise Adoption and OpenRouter Data
Six of OpenRouter's top models are Chinese open-source options. Token processing grew from 2.4 trillion to 47 trillion weekly over one year, driven almost entirely by Chinese models. Companies like Airbnb choose Qwen over ChatGPT for cost and customization.
5. Model Distillation: The Cost Advantage Secret
Chinese labs use distillation—purchasing subscriptions to ChatGPT/Claude, automating prompt generation, and using responses as training data. Technically violates terms of service but legally defensible; even Elon Musk's xAI used distillation for Grok.
6. Why Consumers Prefer Western Models
Consumer markets favor ChatGPT and Gemini due to additional features, voice tools, and integrations rather than token cost. 95% of ChatGPT users rely on free tier. Chinese models win where customers bear token costs directly—enterprise and embedded use cases.
7. The Commoditization Threat
U.S. frontier labs face a bind: they lose consumer money, depend on enterprise token sales, but enterprises will adopt cheaper Chinese alternatives. Continuous development and higher model size/cost creates openings for Chinese labs to quickly catch up and undercut prices.
Key People & Entities
- Sam Altman
- CEO of OpenAI; personal friend of Airbnb founder Brian Chesky
- Brian Chesky
- Founder and CEO of Airbnb; chose Chinese model Qwen over ChatGPT for cost savings
- Elon Musk
- Founder of xAI; admitted using ChatGPT distillation to train Grok AI chatbot in recent trial
- Z.AI
- Chinese AI lab that went public January 2026 on Hong Kong Stock Exchange; develops GLM family of models
- Deepseek
- Chinese AI company that launched competitive open-weight LLM in January 2025, temporarily shocking semiconductor markets
- Alibaba
- Chinese tech giant that developed Qwen, an open-source LLM adopted by Airbnb and other enterprises
- OpenRouter
- Platform helping enterprises manage AI usage across 400+ models; publishes weekly token usage data showing Chinese model dominance
- OpenAI
- U.S. frontier AI lab creator of ChatGPT; losing enterprise market share to cheaper Chinese alternatives
Glossary
- Open-weight LLM
- A large language model where weights (trained parameters) are publicly available for download and local deployment, contrasting with closed-source proprietary models
- Model distillation
- Training technique where a smaller/cheaper model learns from outputs of a larger model by processing its responses as training data
- Token
- Basic unit of text processed by an LLM; pricing typically charged per million tokens for input and output
- On-premise deployment
- AI model customized and fine-tuned for a customer, then installed and run on customer's own internal servers without ongoing licensing fees
- Frontier models
- State-of-the-art AI models representing the latest advances in performance and capability
- API token sales
- Revenue from enterprises using a model's API and paying per number of tokens processed
- Long-horizon benchmarks
- Test cases requiring AI to complete complex multi-step tasks autonomously, such as building an entire website from a single prompt
- Gross profit margin
- Percentage of revenue remaining after subtracting cost of goods sold (direct costs), before operating expenses